Songs & Sounds: A Spotify Story

Data Collection

1. Top Spotify Songs from 2010-2019: This dataset lists the top songs by year across the world and was used in Figures 1, 2, 3, and 4. There are various other variables (such as Title, Artist, Top Genre, and more) and the song rankings are based on Billboard’s rankings. The link for this dataset can be found here.

2. Map Data: This visualization was generated in a multistep process which included 2 open source data sets and one datset from the Spotify API containing data on the availability of songs across the world. The first open source dataset contains country names, 3-letter (Alpha-3) country codes. The second open source dataset contains 2-letter (Alpha-2) country codes.

3. Spotify Dataset: This dataset rates songs on Spotify by different scores for liveliness, energy, loudness, etc. The data is further classified by genre and duration. The link for this dataset can be found here.

4. Personal Spotify Data: Personal streaming data from the past year was requested and collected by one of our group members (Nina) from Spotify. This data was used for Figures 5a, 5b, and 5c.

Data Cleaning

1. Top Spotify Songs from 2010-2019: This data was already clean, but this dataset was also used to generate the location ids for each country that the songs were available for streaming in. We wrote code to generate a new csv that contained the country codes for each song. This csv file was then used in our map-making code, which generated our chloropleth map (Figure-1).

2. Map Data: Since csv originally had Alpha-2 codes, we needed the first open source dataset as an intermediate to convert Alpha-2 codes to Alpha-3 codes. Once this was completed, we were able to merge the dataset with the pre-made geopandas world map coordinate data (via outer join) to create a final dataset containing country name, country geometry, and Alpha-3 codes. This dataset is in geopandas dataframe format, allowing us to plot it onto the world map with color encoding the availability of the top songs by country

3. Spotify Dataset: This data did not need to be cleaned.

4. Personal Spotify Data: The streaming history data that was collected from Spotify initially was in the form of three json files:

2021-02-15 to 2021-08-28

2021-08-28 to 2022-01-06

2022-01-06 to 2022-02-15

These files did not have the specific URIs that were necessary to access the songs from the Spotify API, so we had to write code that would use the song name and the artist name to search through the Spotify API, access the song id, and then attach that song id to the correct song from the json file. The last step was to convert the json file into a csv file, as this would allow us to use plotly.

The resulting csv file for Nina’s Streaming History can be found by clicking on “Nina’s Spotify Dataset”.

The last step was to access information about each song using the Spotify API. The code that executed this step created a csv file that could then be read into a dataframe and used with plotly.

Design Choices

Figure-1. Spotify Access World Map: A chloropleth map is an excellent way to illustrate the geographic range of top songs. Darker colors represent higher concentrations of top songs.

Figure-2. Relationship Between Song Characteristics: This view is used to give a broad overview of the characteristics we are analyzing through a correlational heatmap. The design choices here are limited to the color scheme used (which is a divergent color palette), with the goal being to effectively show differences between attributes with high and low correlation. Readability of axis labels and key labels are also important to note.

Figure-3. Song Chacteristics by Genre Over Time: A Plotly line chart was used, with different colors representing the different attributes. Each data point is marked with a point on the line to show how different genres are only represented in specific years. For instance, the rap genre only has 3 data points, whereas pop has a data point for each year. All characteristics but bpm, decibels, and duration are displayed in this line chart as these characteristics could not be effectively plotted on the same scale. They also are characteristics used to gain a feel for the type of music whereas the other characteristics have more relation to the user experience of the song.

Figure-4. Song Characteristics and Top Artists by Genre:] This view shows the distributions of bpm, decibels, and duration as well as the top artists for each genre. The colors in the legend are used to represent the different graphs; one for each characteristic (bpm, decibels,duration) and then one color for each genre with the top artists plot. The decision to link these two views was made to show what artists and what genres produce which types of music.

Figure-5a. Frequencies of Top 15 Songs: Since we wanted this graph to represent frequencies, a bar graph was the best option. To emphasize that song count is the focus of this graph, we clearly labeled each bar with the frequency, instead of the name of the song. Keeping all the names on the x-axis made the visual very messy (despite angling the text), so we included the names of the songs, as well as their musical features, in the tooltip. We used a continuous single-color scale because this was the most effective way to visually represent the relative counts of each song. A single-color scale is easier to interpret than a diverging one because intuition suggests that darker colors correspond to greater “concentrations”, or frequencies, of songs.

Figure-5b. Comparing Relative Levels of Each Music Feature: If there was more time, we would have wanted to find a way to combine these colors into a image that had levels of concentrations of each color, as depicted on the bar graph. On their own, these colors are simply meant to be eye-catching. A bar graph was also used to show relative levels of each feature, on the same scale.

Figure-5c. Tracking Music Trends Over Time and Summarizing Minimum/Maximum Values of Tempo, Danceability, Energy, and Valence: The purpose of the trend time series is to see how music characteristics change over time, and it uses color to distinguish between music features. The differing scale for each feature warranted the drop down menu, so that users can isolate each feature and see the trend more clearly.

The tree map is an excellent tool for demonstrating the hierarchical structure of:

Feature Name

Min vs Max

Song(s) Name(s)

Dates Played

Reflection

Much of the graph-making done in this project was created in Plotly. There were attempts to make these visualizations using other methods. For instance, the original idea for the Linked View was to use javascript, but we ran across issues with using and manipulaitng a csv file in that context and changed gears. Ultimately, Plotly was the main driver of the viz’s we created because of the aesthetics it provided and its flexibility, which our group preferred strongly over some of the other options.

If starting the project from scratch today, we likely would have tried to use more javascript to make our viz’s more unique. The scope of our project and our goals were such that we did not need this added flexibility to get the job done, and we do think that much of what we created satisfied our original vision. However, knowing what we know now, we could have tried to use js to incorporate audio elements and more uniquely interactive viz’s.

Frequencies of Top 15 Songs

hover over a bar to learn which song it represents!

Figure-5a. Frequencies of Top 15 Songs. The purpose of this visualization is to provide a simple ranking for the top songs that a user was listening to in the selected time period. This graph would allow users to see their top listened-to songs from that time period, and marvel at how obsessively they may have listened to a particular song (ie. Photo ID, in Nina’s case).

By showing the frequencies of the top 15 songs in a user’s listening history, a user is able to see very clearly how much they liked a particular song in comparison to other ones. They are also able to see how their popular songs line up with other people’s favorite songs.

Comparing Relative Levels of Each Music Feature

Figure-5b. Comparing Relative Levels of Each Music Feature. This second bar chart provides a numerical picture of a user’s music taste. While we only have data from one user, it would make sense that users who like similar songs would share similar graphs such as the one above.

Percentages were used, instead of absolute numbers, because not all features use the same scale. Percentages were taken by taking the average for each feature over Nina’s entire streaming history, and dividing that average by the maximum value that the feature could have. Thus, this percentage normalizes each average and allows us to compare one feature to another.

From this graph, there is not one singular feature that overwhelmingly characterizes Nina’s music taste. It appears that danceability, energy, and a slightly major key are the most influential factors for Nina’s music taste.

**INNOVATIVE VIEW: Tracking Music Trends Over Time and Summarizing Minimum/Maximum Values of

Tempo, Danceability, Energy, and Valence** {#figure-5c}

Select which feature you would like to see using the dropdown menu. Use the slider to zoom in on a particular period of time. Use the tree map to find instances of songs who have a minimum/maximum value for a particular feature.

Figure-5c. INNOVATIVE VIEW: Tracking Music Trends Over Time and Summarizing Minimum/Maximum Values of Tempo, Danceability, Energy, and Valence. The time series graph shows how certain characteristics of a user’s music change over time. For example, they will be able to identify the periods of time when their energy is high, and see if that corresponds to a time in their life where they felt energetic and motivated.

This view is innovative because it uses the the tree map to identify the songs who have minimum or maximum values for tempo, energy, danceability, or valence. Once it identifies these songs, it shows the dates that these songs were played. For example, the song with the highest tempo is Anybody But You, played on 2021-09-07 00:28. Accomplishing this task required many steps, such as filtering, grouping, and iterating through rows of data in order to find the songs whose feature values matched the respective minimum/maximum values. Ideally, if someone is unsure of what a feature means, they can click on the feature in the tree map and learn what song is characterized by that feature. If the user knows that song, they then have a better sense of what the time series plot means.

The time series plot is in itself innovative, since it uses time series analysis to represent the moving average trend of each feature. It allows users to see where certain features (such as energy) were the highest, and when they fell. Users can then determine whether these trends in music were correlated to particular emotions/events that happened in their life, making this visual much more personal.

From the three visuals, one can see that the top songs that characterize Nina’s music taste are likely high in energy, danceability, and a major key (Figure-5b. From referencing Figure-5c, the peak of this energy and danceability occurred in September, 2021, right at the beginning of the school year. These features plummeted into finals week, and gradually increased throughout the winter until February. The tree map in Figure-5c also indicates that energy is the most significant feature of Nina’s music taste, as this feature has the greatest number of instances that have the maximum value of energy.